xref: /freebsd/contrib/llvm-project/libcxx/include/__random/poisson_distribution.h (revision a0ca4af9455b844c5e094fc1b09b1390ffa979fc)
1 //===----------------------------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #ifndef _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H
10 #define _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H
11 
12 #include <__config>
13 #include <__random/clamp_to_integral.h>
14 #include <__random/exponential_distribution.h>
15 #include <__random/is_valid.h>
16 #include <__random/normal_distribution.h>
17 #include <__random/uniform_real_distribution.h>
18 #include <cmath>
19 #include <iosfwd>
20 #include <limits>
21 
22 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
23 #  pragma GCC system_header
24 #endif
25 
26 _LIBCPP_PUSH_MACROS
27 #include <__undef_macros>
28 
29 _LIBCPP_BEGIN_NAMESPACE_STD
30 
31 template <class _IntType = int>
32 class _LIBCPP_TEMPLATE_VIS poisson_distribution {
33   static_assert(__libcpp_random_is_valid_inttype<_IntType>::value, "IntType must be a supported integer type");
34 
35 public:
36   // types
37   typedef _IntType result_type;
38 
39   class _LIBCPP_TEMPLATE_VIS param_type {
40     double __mean_;
41     double __s_;
42     double __d_;
43     double __l_;
44     double __omega_;
45     double __c0_;
46     double __c1_;
47     double __c2_;
48     double __c3_;
49     double __c_;
50 
51   public:
52     typedef poisson_distribution distribution_type;
53 
54     _LIBCPP_HIDE_FROM_ABI explicit param_type(double __mean = 1.0);
55 
56     _LIBCPP_HIDE_FROM_ABI double mean() const { return __mean_; }
57 
58     friend _LIBCPP_HIDE_FROM_ABI bool operator==(const param_type& __x, const param_type& __y) {
59       return __x.__mean_ == __y.__mean_;
60     }
61     friend _LIBCPP_HIDE_FROM_ABI bool operator!=(const param_type& __x, const param_type& __y) { return !(__x == __y); }
62 
63     friend class poisson_distribution;
64   };
65 
66 private:
67   param_type __p_;
68 
69 public:
70   // constructors and reset functions
71 #ifndef _LIBCPP_CXX03_LANG
72   _LIBCPP_HIDE_FROM_ABI poisson_distribution() : poisson_distribution(1.0) {}
73   _LIBCPP_HIDE_FROM_ABI explicit poisson_distribution(double __mean) : __p_(__mean) {}
74 #else
75   _LIBCPP_HIDE_FROM_ABI explicit poisson_distribution(double __mean = 1.0) : __p_(__mean) {}
76 #endif
77   _LIBCPP_HIDE_FROM_ABI explicit poisson_distribution(const param_type& __p) : __p_(__p) {}
78   _LIBCPP_HIDE_FROM_ABI void reset() {}
79 
80   // generating functions
81   template <class _URNG>
82   _LIBCPP_HIDE_FROM_ABI result_type operator()(_URNG& __g) {
83     return (*this)(__g, __p_);
84   }
85   template <class _URNG>
86   _LIBCPP_HIDE_FROM_ABI result_type operator()(_URNG& __g, const param_type& __p);
87 
88   // property functions
89   _LIBCPP_HIDE_FROM_ABI double mean() const { return __p_.mean(); }
90 
91   _LIBCPP_HIDE_FROM_ABI param_type param() const { return __p_; }
92   _LIBCPP_HIDE_FROM_ABI void param(const param_type& __p) { __p_ = __p; }
93 
94   _LIBCPP_HIDE_FROM_ABI result_type min() const { return 0; }
95   _LIBCPP_HIDE_FROM_ABI result_type max() const { return numeric_limits<result_type>::max(); }
96 
97   friend _LIBCPP_HIDE_FROM_ABI bool operator==(const poisson_distribution& __x, const poisson_distribution& __y) {
98     return __x.__p_ == __y.__p_;
99   }
100   friend _LIBCPP_HIDE_FROM_ABI bool operator!=(const poisson_distribution& __x, const poisson_distribution& __y) {
101     return !(__x == __y);
102   }
103 };
104 
105 template <class _IntType>
106 poisson_distribution<_IntType>::param_type::param_type(double __mean)
107     // According to the standard `inf` is a valid input, but it causes the
108     // distribution to hang, so we replace it with the maximum representable
109     // mean.
110     : __mean_(isinf(__mean) ? numeric_limits<double>::max() : __mean) {
111   if (__mean_ < 10) {
112     __s_     = 0;
113     __d_     = 0;
114     __l_     = std::exp(-__mean_);
115     __omega_ = 0;
116     __c3_    = 0;
117     __c2_    = 0;
118     __c1_    = 0;
119     __c0_    = 0;
120     __c_     = 0;
121   } else {
122     __s_        = std::sqrt(__mean_);
123     __d_        = 6 * __mean_ * __mean_;
124     __l_        = std::trunc(__mean_ - 1.1484);
125     __omega_    = .3989423 / __s_;
126     double __b1 = .4166667E-1 / __mean_;
127     double __b2 = .3 * __b1 * __b1;
128     __c3_       = .1428571 * __b1 * __b2;
129     __c2_       = __b2 - 15. * __c3_;
130     __c1_       = __b1 - 6. * __b2 + 45. * __c3_;
131     __c0_       = 1. - __b1 + 3. * __b2 - 15. * __c3_;
132     __c_        = .1069 / __mean_;
133   }
134 }
135 
136 template <class _IntType>
137 template <class _URNG>
138 _IntType poisson_distribution<_IntType>::operator()(_URNG& __urng, const param_type& __pr) {
139   static_assert(__libcpp_random_is_valid_urng<_URNG>::value, "");
140   double __tx;
141   uniform_real_distribution<double> __urd;
142   if (__pr.__mean_ < 10) {
143     __tx = 0;
144     for (double __p = __urd(__urng); __p > __pr.__l_; ++__tx)
145       __p *= __urd(__urng);
146   } else {
147     double __difmuk;
148     double __g = __pr.__mean_ + __pr.__s_ * normal_distribution<double>()(__urng);
149     double __u;
150     if (__g > 0) {
151       __tx = std::trunc(__g);
152       if (__tx >= __pr.__l_)
153         return std::__clamp_to_integral<result_type>(__tx);
154       __difmuk = __pr.__mean_ - __tx;
155       __u      = __urd(__urng);
156       if (__pr.__d_ * __u >= __difmuk * __difmuk * __difmuk)
157         return std::__clamp_to_integral<result_type>(__tx);
158     }
159     exponential_distribution<double> __edist;
160     for (bool __using_exp_dist = false; true; __using_exp_dist = true) {
161       double __e;
162       if (__using_exp_dist || __g <= 0) {
163         double __t;
164         do {
165           __e = __edist(__urng);
166           __u = __urd(__urng);
167           __u += __u - 1;
168           __t = 1.8 + (__u < 0 ? -__e : __e);
169         } while (__t <= -.6744);
170         __tx             = std::trunc(__pr.__mean_ + __pr.__s_ * __t);
171         __difmuk         = __pr.__mean_ - __tx;
172         __using_exp_dist = true;
173       }
174       double __px;
175       double __py;
176       if (__tx < 10 && __tx >= 0) {
177         const double __fac[] = {1, 1, 2, 6, 24, 120, 720, 5040, 40320, 362880};
178         __px                 = -__pr.__mean_;
179         __py                 = std::pow(__pr.__mean_, (double)__tx) / __fac[static_cast<int>(__tx)];
180       } else {
181         double __del = .8333333E-1 / __tx;
182         __del -= 4.8 * __del * __del * __del;
183         double __v = __difmuk / __tx;
184         if (std::abs(__v) > 0.25)
185           __px = __tx * std::log(1 + __v) - __difmuk - __del;
186         else
187           __px = __tx * __v * __v *
188                      (((((((.1250060 * __v + -.1384794) * __v + .1421878) * __v + -.1661269) * __v + .2000118) * __v +
189                         -.2500068) *
190                            __v +
191                        .3333333) *
192                           __v +
193                       -.5) -
194                  __del;
195         __py = .3989423 / std::sqrt(__tx);
196       }
197       double __r  = (0.5 - __difmuk) / __pr.__s_;
198       double __r2 = __r * __r;
199       double __fx = -0.5 * __r2;
200       double __fy = __pr.__omega_ * (((__pr.__c3_ * __r2 + __pr.__c2_) * __r2 + __pr.__c1_) * __r2 + __pr.__c0_);
201       if (__using_exp_dist) {
202         if (__pr.__c_ * std::abs(__u) <= __py * std::exp(__px + __e) - __fy * std::exp(__fx + __e))
203           break;
204       } else {
205         if (__fy - __u * __fy <= __py * std::exp(__px - __fx))
206           break;
207       }
208     }
209   }
210   return std::__clamp_to_integral<result_type>(__tx);
211 }
212 
213 template <class _CharT, class _Traits, class _IntType>
214 _LIBCPP_HIDE_FROM_ABI basic_ostream<_CharT, _Traits>&
215 operator<<(basic_ostream<_CharT, _Traits>& __os, const poisson_distribution<_IntType>& __x) {
216   __save_flags<_CharT, _Traits> __lx(__os);
217   typedef basic_ostream<_CharT, _Traits> _OStream;
218   __os.flags(_OStream::dec | _OStream::left | _OStream::fixed | _OStream::scientific);
219   return __os << __x.mean();
220 }
221 
222 template <class _CharT, class _Traits, class _IntType>
223 _LIBCPP_HIDE_FROM_ABI basic_istream<_CharT, _Traits>&
224 operator>>(basic_istream<_CharT, _Traits>& __is, poisson_distribution<_IntType>& __x) {
225   typedef poisson_distribution<_IntType> _Eng;
226   typedef typename _Eng::param_type param_type;
227   __save_flags<_CharT, _Traits> __lx(__is);
228   typedef basic_istream<_CharT, _Traits> _Istream;
229   __is.flags(_Istream::dec | _Istream::skipws);
230   double __mean;
231   __is >> __mean;
232   if (!__is.fail())
233     __x.param(param_type(__mean));
234   return __is;
235 }
236 
237 _LIBCPP_END_NAMESPACE_STD
238 
239 _LIBCPP_POP_MACROS
240 
241 #endif // _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H
242